Diagnostic value of cerebrospinal fluid tau, neurofilament, and progranulin in definite frontotemporal lobar degeneration

The study population consisted of definite FTLD patients (n = 46), defined by genetic carrier status and/or postmortem neuropathological confirmation [27, 28]. These patients could be subdivided into FTLD-TDP (n = 34, including 19 FTLD-C9orf72, 9 FTLD-GRN, 1 FTLD-VCP, and 1 FTLD-TBK1 symptomatic mutation carriers; 19 pathologically confirmed), FTLD-tau (n = 10, including 1 FTLD-MAPT symptomatic mutation carrier; 9 pathologically confirmed), or FTLD-other (n = 2, both pathologically confirmed). Clinical diagnosis of definite FTLD patients consisted of the behavioral variant frontotemporal dementia (n = 30), AD (n = 7), primary progressive aphasia (n = 3), progressive supranuclear palsy (n = 2), corticobasal degeneration (n = 1), and other types of dementia (n = 3). AD patients (n = 45) were clinically diagnosed based on IWG-2 criteria, including the AD CSF biomarker panel (pathological cut-offs: Aβ_1–42 < 638.5 pg/mL, t-tau > 296.5 pg/mL, p-tau₁₈₁ > 56.5 pg/mL) [10, 29]. Together with extensive clinical follow-up (median 4.9 (range 2.7–7.9) years) this added to the diagnostic certainty. The control group (n = 20) consisted of elderly people who had no neurological or psychiatric antecedents and no organic disease involving the central nervous system following extensive clinical examination (patients with polyneuropathy, n = 8; patients with subjective complaints, n = 12). All CSF samples (full cohorts) and available blood samples (controls, n = 18; AD, n = 42; FTLD, n = 30) were selected from the Institute Born-Bunge (IBB) Biobank, Antwerp, Belgium [30]. Data including gender, Mini-Mental State Examination (MMSE) score, age at time of CSF sampling, age at onset, and age at death (if applicable) were available for the majority of the patients. This study was approved by the ethics committee of the University of Antwerp, Antwerp, Belgium (B300201420405).

Lumbar puncture (LP), and CSF and blood sampling and handling was performed according to a standardized protocol [30, 31]. All CSF and blood samples were stored at the IBB Biobank in polypropylene vials at −80 °C until analysis.

CSF biomarker levels were quantified using commercially available single-analyte enzyme-linked immunosorbent assay (ELISA) kits (one kit lot each), strictly following the manufacturer’s instructions (INNOTEST β-Amyloid(_1–42), INNOTEST hTau-Ag, and INNOTEST Phospho-Tau(_181P) from Fujirebio Europe, Belgium; Nf-light from UmanDiagnostics, IBL International GmbH, Germany; pNF-H V2 from EnCor Biotechnology Inc., USA; human progranulin from Adipogen Inc., Korea). The last kit was also used to quantify serum levels of progranulin. All samples were run in duplicate, blinded for diagnosis. Intra-assay coefficient of variation was below 10% for all analytes.

Statistical testing was performed using IBM SPSS Statistics 23, GraphPad Prism 6, and the R package ‘pROC’ in RStudio (version 1.0.136) [32]. As some variables were not normally distributed and numbers in FTLD subgroups were small, nonparametric analyses were used. Kruskal-Wallis analyses were performed to describe the dementia patient cohorts and compare biomarker levels between groups. Post-hoc analyses included Dunn’s correction for multiple comparisons. For pairwise comparisons between FTLD pathological and genetic subgroups, Mann-Whitney U tests were performed. Categorical variables were analyzed with a Chi-square test. Spearman’s ρ was calculated to determine correlations. Logistic regression models were generated using a forward selection method of AD biomarkers alone, or of all available single CSF biomarkers. For these analyses, biomarker data were log₁₀-transformed to achieve normality. Receiver operating characteristic (ROC) curve analyses were used to obtain area under the curve (AUC) values with 95% confidence intervals (CIs) for differentiation between groups [32, 33]. AUC values were compared using DeLong tests. The maximal sum of sensitivity and specificity (maximized Youden’s index) was calculated to determine cut-off values for progranulin. For all analyses, p values below 0.05 were considered statistically significant.

Demographic, clinical, and biomarker data for all groups are summarized in Table 1. Two FTLD patients were excluded from the statistical analysis as none of their AD biomarker values could be determined, probably related to preanalytical factors. Biomarker values outside of the assay limits of detection were set to the lowest/highest detection point ±20%, and this value was used in nonparametric statistical analysis (t-tau: 1 FTLD and 5 AD patients; p-tau₁₈₁: 3 FTLD patients; Nf-L: 3 FTLD patients, 2 controls, and 3 AD patients; pNf-H: 9 FTLD, 7 controls, and 8 AD patients). There was no difference in biomarker levels between males and females (p = 0.53). Age at LP correlated significantly with Nf-L levels in controls and AD patients, but not in the FTLD group (controls: ρ = 0.837, p < 0.001; AD: ρ = 0.415, p < 0.01; FTLD: ρ = 0.006, p = 0.97). MMSE score correlated with Aβ_1–42 (0.436, p < 0.01) and Nf-L (−0.457, p < 0.01) levels in AD patients. No notable correlations were found for the different clinical features and biomarker levels in the FTLD group or its subgroups.

To differentiate between FTLD and AD, a reference logistic regression model generated with the established AD biomarkers achieved an AUC of 0.942 (95% CI 0.891–0.981). The only investigated biomarker that added value to this model was Nf-L, resulting in an AUC of 0.948 (95% CI 0.900–0.984) but this improvement was not significant (p = 0.30). The addition of age as a covariate did not change the result of the logistic model (AUC 0.951, 95% CI 0.908–0.956; p = 0.70).

For the differentiation between FTLD and controls, Aβ_1–42 and Nf-L were the best predictors with an AUC of 0.881 (95% CI 0.761–0.971). The addition of age as a covariate did not significantly alter this model, gaining an AUC of 0.879 (95% CI 0.759–0.970; p = 0.66).

In this study we confirm the use of decreased progranulin levels in either serum or CSF as a marker for mutation status in FTLD-GRN patients. No evidence was found that progranulin levels in CSF were significantly altered in FTLD patients without GRN mutations.

There are contradicting results about the correlation of blood and CSF progranulin levels in control subjects, being either absent [37, 38] or present [16, 17]. We did not find a significant correlation in either controls or AD patients, but CSF and serum progranulin levels did correlate very strongly in our FTLD-GRN subgroup (ρ = 0.821). This has only been looked at in one other study so far and, while the correlation was also significant, it was not as strong (Pearson’s r = 0.54 [38]). In our entire cohort with both serum and CSF available (n = 87), about 20.7% of CSF progranulin variability could be explained by serum progranulin variability (ρ² = 0.207). This is considerably higher than previous studies which found that plasma progranulin variability only explained 6.2% or 13.1% of CSF progranulin variability, respectively [16, 17]. However, these studies did not include GRN mutation carriers in their analysis, and when we exclude these subjects the Spearman’s ρ² was indeed only 0.099 (i.e., 9.9% explained) in our entire noncarrier group (n = 80). Another difference is the use of the nonparametric Spearman’s correlation coefficient in this study instead of the Pearson’s correlation coefficient. Using the latter in our entire cohort we found an r² of only 0.105 (i.e., 10.5% explained). Focusing on the subgroup of GRN mutation carriers with available serum and CSF (n = 7), we found that 67.4% of CSF progranulin variability could be explained by serum progranulin variability. This is much higher than the 29% that was reported recently [38]. Our findings indicate that progranulin might thus be assessed in serum instead of CSF, but only for the FTLD-GRN group. As patient cohorts were small in both studies, further investigation of this relationship is definitely warranted.

In the entire FTLD group there was a marked variability in Aβ_1–42 and t-tau levels. The alteration in Aβ_1–42 was not related to specific FTLD subtypes, although it was more prominent in FTLD-tau. In some patients, a low level of Aβ_1–42 could be attributed to an Aβ copathology and/or APOE ε4 carrier status, but it might also be connected to individual variation in the Aβ production process [39].

The variability of t-tau was an interesting finding, as the main purpose in using the established AD markers was to see if tau proteins had an additional value in FTLD subtypes, in particular for FTLD-tau. However, neither t-tau nor p-tau₁₈₁ were significantly different between FTLD-TDP and FTLD-tau, or between FTLD patients and controls. We found that the neuropathological subtypes also had comparable p-tau₁₈₁/t-tau ratios (although slightly lower in FTLD-TDP than FTLD-tau). This contradicts previous studies that stated that the ratio was significantly lower in FTLD-TDP [21, 22, 40]. It should be noted that, while they gained significance, differences in p-tau₁₈₁/t-tau ratios in these previous studies were also very small. Additionally, there has been much discussion about the (unspecific) pathological changes that would result in a differential p-tau₁₈₁/t-tau ratio between FTLD-TDP and FTLD-tau [40]. In fact, the ratio has been contradictorily decreased either by lower p-tau₁₈₁ or higher t-tau levels [21, 41]. As such, and together with our data in pathologically confirmed patients, we doubt that the p-tau₁₈₁/t-tau ratio will be valuable in the diagnosis of FTLD subgroups. Interestingly, when looking at genetic subtypes of FTLD-TDP, it appears that t-tau is specifically elevated in patients carrying a GRN mutation since their levels were significantly higher than those of C9orf72 mutation carriers, at least explaining some of the variability in the FTLD group (see also the paragraph ‘Differences within FTLD subtypes’ below).

With regard to neurofilament levels in controls, AD, and FTLD subjects, there was only a significant increase in Nf-L in FTLD. To evaluate the influence of amyotrophic lateral sclerosis (ALS), which is characterized by high Nf-L levels [42], statistical analysis was performed without FTLD-TDP patients with associated ALS (n = 5) which had no effect on the results. In comparison to AD we observed a 2.5-fold increase in FTLD, which is in concordance with other studies that found a two- to threefold concentration of Nf-L in FTLD compared with AD [26, 40, 43]. Some older studies did not find a significant difference between the two types of dementias, but these studies had the limitation of assessing clinical FTLD patients only and separating the AD cohort based on age of onset [25, 44]. We observed more than a threefold increase in Nf-L levels in FTLD compared with controls, which is again in agreement with other studies [25, 26, 43, 45]. As for pNf-H, previous studies with clinical patient cohorts have reported differences in pNf-H levels between FTLD and controls, but differentiation with AD has not been possible [25, 44, 46]. By confirming the lack of significant differences in definite FTLD patients, we conclude that focus should remain on Nf-L rather than pNf-H.

While we established that Nf-L was higher in FTLD in general, we investigated if there was an association with a specific underlying pathology and found a trend towards increased Nf-L values in FTLD-TDP compared with FTLD-tau (p = 0.052). As the same relation has gained significance in other studies [26, 40] the small number of FTLD-tau patients in our study likely limited our findings. In the genetic subgroups, a significant difference for Nf-L was found between GRN and C9orf72 mutation carriers (see also the paragraph ‘Differences within FTLD subtypes’ below).

In this study, the diagnostic accuracy obtained with the established AD biomarkers was used as a reference for comparison with the other investigated biomarkers. It should be noted that, while the use of AD CSF biomarkers is known to increase diagnostic certainty, the clinical diagnosis of AD patients remains suboptimal which is a limitation of this study and has likely influenced differential diagnostic accuracy. Our results show that only p-tau₁₈₁ and Aβ_1–42 were necessary to differentiate between AD and FTLD, confirming previous results in definite FTLD patients [47, 48]. The discriminative power could be improved on with the inclusion of Nf-L. While diagnostic accuracies of AD markers on their own and Nf-L on its own have been described previously, only one study performed a joint ROC analysis and also reported an improved diagnostic accuracy when using the combination of Aβ_1–42, p-tau₁₈₁, and Nf-L in comparison with Aβ_1–42 and p-tau₁₈₁ alone [25]. Nf-L also showed added value in the differentiation between FTLD and controls, confirming its usefulness as a marker for FTLD [40, 45].

As stated above, both CSF levels of t-tau and Nf-L were markedly higher in FTLD-GRN patients compared with other FTLD subgroups (Table 1). There are limited data available with which to compare these findings, but for Nf-L at least it has been described that levels are significantly higher in FTLD-GRN than in FTLD-C9orf72 and FTLD-MAPT [45]. Regarding tau, one study in contrast reported lower levels of t-tau and p-tau₁₈₁ in FTLD patients with a GRN mutation than FTLD patients without a GRN mutation, although they were not expecting this result [49]. Indeed, there is evidence that FTLD-GRN patients would specifically have more generalized brain atrophy, more white matter lesions, a faster rate of neurodegeneration, and a resulting shorter disease duration [50‐53]. Our findings certainly support this notion, as the common feature of both t-tau and Nf-L proteins is that they are markers for general neurodegeneration.